1. Wind energy in Europe under future climate conditions.
The statistical downscaling of a CMIP5 model ensemble
Abstract
As the climate system is changing and fossil resources are diminishing, wind has emerged as an
important source of renewable energy. However the amount of energy in the wind is sensitive to
changes in wind speed. Projections of future large-scale circulations over Europe indicate potential
changes, but it is still unknown if/how these changes might affect the wind energy availability over
Europe. Estimations of changes in wind energy availability are important in order to assist the
development of wind power industries. Projections of future climate and large-scale circulation
systems are commonly performed using global climate models (GCMs). However to study the effect
of a change in large-scale circulation systems on the wind power availability, the large-scale
information of the GCMs needs to be downscaled to higher resolutions.
The overall objective of this dissertation is to investigate the impact of climate change on the future
wind power potential over Europe, using an ensemble of statistically downscaled CMIP5 Earth
System models (ESMs). For this purpose, the dissertation outlines four major parts: i) the
development of a statistical downscaling method, ii) the evaluation of the method, iii) the
evaluation of its driving GCMs (the ESMs) and iv) the application of the method.
In the first part, the approach to downscale the large-scale information to the small-scale wind
speed climatology at hub-height is developed. This is done for a specific site in the Netherlands
(Cabauw) using long-term wind speed observations from a measurement mast, reanalysis data from
ERA-Interim andlarge-scale model data from ECHAM5. The method statistically relates the large-
scale data (the predictors) to the small-scale hub-height wind speed climate via a regression transfer
function. In contrast to common downscaling methods, this dissertation develops a method based
on the parameters of the probability density functions (PDFs), for different stability conditions
separately and includes a variable evaluation prior to the selection of the predictors. The regression
results indicate that the seasonal and diurnal conditions of the atmospheric boundary layer are
important in defining which large-scale variables are best in predicting the small-scale wind
climatology. During wintertime the large-scale dynamics typically dominate the near-surface wind
speeds, hence ECHAM5 is skilful in representing the hub-height wind speeds and little improvement
can be brought by the statistical downscaling. On the other hand, during summer, ECHAM5 is not
skilful in representing the hub-height wind speed PDF due to the rather local character of the
summertime winds. However, the regression analysis shows that during convective summer day
conditions the observed hub-height wind speed is strongly linked to the wind speed at higher,
skilfully represented levels. The summer-day hub-height wind speed PDF can therefore skilfully be
predicted by the wind speedPDF parameters at 500m as the only predictors. However this is not the
case during summer nights. During these very stable conditions, the boundary layer is much more
1

2. shallow and the regression analysis indicates that the addition of information on the temperature
gradient between the ABL and above substantially improves the simulation of the observed hub-
height wind speed PDFs.
A second part evaluates whether the statistical models, which are developed in Cabauw, can be
used at other locations in Europe. The comparison of the downscaled winds with observed near-
surface winds over Europe shows that the spatial extent of the regions in which the downscaling
models are capable in representing observed hub-height winds, depends on the diurnal, the
seasonal and the local conditions (like the orography and the presence of regional wind systems).
Depending on the season and time of the day, regions in Europe are defined for which the
downscaling model can be skilfully applied.
Before the downscaling method is applied on an ensemble of ESMs, the ESMs are evaluated on their
representation of the predictors (part 3). Predictors of statistical downscaling models are commonly
derived from upper-atmospheric fields, because these variables are likely to be better represented
by thelarge-scale models. Since the regression analysis on ECHAM5 (step 1) indicates that the near-
surface fields provide the best predictors, the ESMs are evaluated on their representation of wind
and temperature PDFs in the lower 1.5km of the atmosphere, using ERA-Interim as the reference.
This height-dependent evaluation approach of the ESMs is extra relevant since, compared to former
generation GCMs, the resolution of the ESMs is high and their representation of the land-
atmosphere interac on processes are described in greater detail. The results show that the ESMs s
wind speed and temperature variables of the lower 1.5km are suitable to drive statistical
downscaling models over most of Europe. However some small-scale and large-scale biases are
present. Above coastal bays and capes, small-scale biases in the ESMs result in unskilful wind speed
PDFs up to 600m. Orography might affects wind speeds throughout the lowest 1.5km of the
atmosphere. This is mostly the case during summer and daytime conditions. During winter, the
small-scale biases propagate less high. With exception of the biases at the small scale, the surface
wind speed PDFs north of 45°N are well represented by all the ESMs. South of 45°N, winds are
affected by a large-scale bias originating from errors in the representation of the large-scale
circulation, especially during winter. The large-scale wind bias is suggested to be related with a
largely exaggerated latitudinal pressure gradient, leading to the too strong westerlies in the
Northern Hemisphere mid-latitudes. On the other hand, the representation of the temperature PDF
by the ESMs is slightly less affected by biases acting at the small scale. However a large-scale
temperature PDF bias, related to too cold temperatures, is present over the North Atlantic Ocean
and the east of Europe exhibits temperatures that are too high in summer. Most indentified large-
scale biases are independent from height and therefore also adopted by downscaling models
whichare based on upper-atmospheric fields, underlining the importance of model evaluation
before downscaling.
The last part performs the statistical downscaling method using only the skilful ESM fields as
possible predictors and focusing only at the regions where the downscaling has shown to be skilful.
Hub-heightwind speeds are downscaled for three periods: present-day (1989-2000), near future
2

3. (2020-2049) and end of the century period (2070-2099). The hub-height wind speed PDF
parameters are converted into power for a sample turbine and the change in power relative to the
present-day climate is analyzed in a Bayesian ensemble approach. This probabilistic approach
weights the participation of the ESMs in the ensemble on their bias and convergence. The analysis
exhibits the importance of the PDF based approach. It shows that in a climate in which both PDF
parameters increase (resulting in a wider and more symmetric PDF) the true power output will
increase to a relatively lesser extent than expected from the change in mean wind speed. This is for
example the case in Western-Europe during wintertime, where the expected change in large-scale
westerly wind speeds might lead to an (insignificant) increase of power output of about 5%. The
opposite situation is true for the Mediterranean region, where the decreasing PDF parameters
(resulting in a more narrow and skewed PDF) have a large effect on the power output. In these
southern regions, power outputs are expected to decrease (significantly) by magnitudes up to 16%.
When power change estimations would be calculated from the change in the mean of the wind (as
itis common practice), the expected decrease in the Mediterranean wind power output would be
overestimated by almost 20% of the change. These inaccuracies in the estimations of the change in
power output resulting from neglecting the changes in the PDF (its skewness and width) of the wind,
are in the order of magnitude relevant for future wind power yield estimations. Not only the
changes in the PDF of the wind speed, but also the form of the Cp-curve has shown to affect the
change in power output, although this is not significant.
Finally a comparison of the changes in power output projections with and without the statistically
downscaling step, indicates no substantially differences. Although the present-day summertime
hub-height wind speed PDF is clearly improved by the downscaling, the future change of is not
affected by thedownscaling practice. This implies that at least for wind power, and possible other
applications, the direct output of the current generation of the ESMs do not necessarily need
downscaling.
3